Show Table of Contents
5.3.3. Setting Concurrency Limits
JBoss ON can handle large numbers of agents, potentially hundreds. The JBoss ON server can possibly be flooded with messages if many agents attempt to communicate with the server simultaneously. This can happen if the JBoss ON server is restarted after being down for a period of time; when JBoss ON agents detect that the JBoss ON server has come back, they all immediately attempt to send it a backlog of messages.
The JBoss ON server can have a configurable limit on the number of concurrent messages that can be processed at one time, to mitigate any risk of flooding the server. Any messages that come in past that limit are dropped and the agent is asked to send them later.
All of the concurrency-related parameters are listed in Table 10, “rhq-server.properties Parameters for Concurrency Limits”.
Concurrency limits not only limit the number of agent connections, but also the number of connections to the GUI and other web connections to the server. There are three primary parameters that control the concurrency limits:
- A global limit on the total number of incoming messages to the server (
rhq.communications.global-concurrency-limit).This is the total number of allowed agent connections. There are other concurrency limits for specific message types which can help tune performance for content downloads, inventory synchronization, and other resource-intensive or recurring agent operations. Those concurrency limits apply only to those specific message types, and those limits are evaluated independently of each other. The global concurrency limit is the total cap for all agent connections. This is the effective concurrency limit, even if the sum of the other concurrency limits is higher.
- A limit on the total number of concurrent web connections allowed (
rhq.server.startup.web.max-connections).This counts any client connection which connects to the JBoss ON server over an HTTP or HTTPS connection. This includes web GUI connections, of course, but it also includes all agent connections which use the (default)
ssslservlettransports.The limit on web connections is the same for both non-secured HTTP requests and HTTPS requests, but the limit is additive so HTTP and HTTPS connections count against different pools. The total maximum connections allowed is actually twice whatever the
rhq.server.startup.web.max-connectionsvalue is. For example, if the setting is 300, then 300 HTTP requests are allowed and 300 HTTPS requests are allowed, for total of 600 concurrent web connections.
- Limits on the number of downloads from agents (
rhq.server.agent-downloads-limit) and from other clients (
Example 6. Concurrency Limits
rhq.server.startup.web.max-connections=200 rhq.server.agent-downloads-limit=45 rhq.server.client-downloads-limit=5 rhq.communications.global-concurrency-limit=30
Table 10. rhq-server.properties Parameters for Concurrency Limits
|rhq.server.startup.web.max-connections|| Sets a limit on the number of web connections that can be concurrently created, including both connections to the GUI and connections by agents.
The limit on web connections is the same for both HTTP and HTTPS (secure) requests, so the total max connections allowed is actually twice what this setting is. For example, if the max web connections is set to 300, then 300 HTTP requests will be allowed and 300 HTTPS requests will be allowed, for a total of 600 concurrent web connections.
If agent requests are routed over web connections, make sure that the
Sets the total number of agent messages that come into the server. This only affected incoming agent messages, not GUI requests. If this global concurrency limit is set to 300, no more than 300 total agent messages can be processed at any one time, regardless of what kinds of messages are coming in.
Even if the sum of the other concurrency limits are higher than this global limit, they are capped at this global limit since there can never be more messages processed than the global limit.
This value should be slightly lower than the number of allowed web connections so that web connections to the GUI are not blocked when there is a high agent load.
|rhq.server.concurrency-limit.inventory-report||Inventory reports are sent from the agent when the agent starts up, and periodically thereafter. Inventory reports can be large, depending on the number of resources on the agent machine.|
|rhq.server.concurrency-limit.availability-report||Availability reports are regularly sent from the agent, typically every 60 seconds. Availability reports are usually very small, but occur in large numbers due to the high frequency of their transmission.|
|rhq.server.concurrency-limit.inventory-sync||Inventory synchronizations occur when the agent needs to synchronize its inventory with that of the server. Agents typically synchronize at startup. Traffic that flows as part of inventory synchronizations is usually large, depending upon the number of resources managed by the agent.|
|rhq.server.concurrency-limit.content-report||Content reports are similar to inventory reports except they contain information about discovered content (i.e., installed packages of software). These reports can be large depending on the number of installed software the agent has discovered and is managing.|
|rhq.server.concurrency-limit.content-download||Content downloads occur when a resource on an agent needs to ask for the content of a package version, usually for the purpose of installing the package.|
|rhq.server.concurrency-limit.measurement-report||Measurement reports are periodically sent to the server whenever the agent completes measurement collections. The number and size of measurement reports can vary, depending on the number and frequency of measurements scheduled to be collected. The greater the number of schedule measurements the agent needs to collect, the more frequently measurement reports are sent, and the larger the reports will be.|
|rhq.server.concurrency-limit.measurement-schedule-request||Similar to inventory synchronization, measurement schedule requests are sent to the agent asking the server for an up-to-date set of measurement schedules that have to be collected.|